I need to find the temperature distribution of a square plate using the Laplace equation by using FDM:

$$ \frac{d^2T}{dx^2} + \frac{d^2T}{dy^2} = 0$$

But there is a heat flux entering from the top boundary $q = 10^5 W/m^2$. How do I incorporate this heat flux into the problem? Should I:

  1. Add it as a boundary condition by considering area $A=1$ and make the top boundary $10^5$? I saw something of this sort in an FVM textbook but dimensionally, it doesn't make sense to me.

  2. Consider the 1-D heat flux equation:

$$ q = -k \frac{dT}{dy} $$

and discretize it as $$\frac{T_{i} - T_{i-1}}{dy} = -q/k$$

and incorporate this into every x-iteration?

This method makes sense to me but when I tried it, the contour plot showed the temperature in negatives as so: enter image description here

If I am adding a heat flux $10^5$, why is the temperature in negatives?

  1. Another method?
  • 1
    $\begingroup$ At top boundary, $q = -k \nabla T \cdot n > 0$, then heat is flowing out of the top boundary. You have $T=0$ on other three sides, so thinking physically, negative temperature is not wrong in this case. $\endgroup$
    – cfdlab
    Mar 8, 2021 at 7:24
  • 1
    $\begingroup$ You can use ghost points, in this case. That is, you introduce (conceptually) an additional row of points $\Delta y$ above the first one, and you compute the values of the temperature along this line such that a first-order discretisation of the heat flux (using these ghost points and the points of the first row) yields the prescribed flux $\phi$. For instance ($i$ is the index along $x$, $j$ for $y$): $\pm\lambda\frac{T_{ghost~up,j} - T_{i,j}}{\Delta y} = \phi_j$, from which you can compute $T_{ghost~up,j}$ and use it to compute the discrete laplacin in the first (real) row of points. $\endgroup$
    – Laurent90
    Mar 8, 2021 at 8:33
  • 1
    $\begingroup$ Another possibility that may be easier to implement is to put a source term in the guard cells, and enforce zero gradient of the temperature on the external sides of the guard cells. $\endgroup$ Mar 8, 2021 at 9:17
  • 1
    $\begingroup$ Your question is confusing the modeling part and the numerics. First derive a model that consists of the PDE and its boundary conditions. Then you can wonder how to discretize it. $\endgroup$ Mar 8, 2021 at 9:47
  • $\begingroup$ Thanks everyone for your inputs. I was able to solve it. @WolfgangBangerth yes the PDE is the Laplace equation and the boundary conditions are zero on all sides. $\endgroup$
    – justauser
    Mar 8, 2021 at 11:58

1 Answer 1


Because a heat flux has a direction, and from what you are describing, you are adding a heat flux in the normal direction -- out of the domain. So if you started with a constant temperature plane and they draw heat out of it, you'd end up with a negative temperature.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.